Human-verification tablet next to the press

The problem

Nobody wants AI writing resin batches into the ERP unchecked.

The main cultural blocker to digitization is the fear that AI will enter data on its own and contaminate the ERP. But technicians don't want to type either. Every iLEAN capture — the photographed order, the voice-dictated incident, the supplier email — shows up on a tablet as a visual summary.

• The main cultural brake on digitization in a closure plant is the fear that AI will enter data on its own and someone will have to clean up the ERP master afterwards.
• It is a reasonable fear: a wrong resin batch or color code in the master travels to the batch record, the certificate of conformity and the bottler's incoming inspection. Correcting it afterwards leaves doubts about every record around it.
• But technicians do not want to type either. They are there to keep presses, molds and color changes running, and every form added at the press is filled in late or not at all.
• So plants stay stuck between unvalidated data contaminating the ERP and data never captured out of distrust. Neither supports the batch-to-batch traceability bottlers audit.

How it fits the IRIS system

Connect with human verification at the press — two taps, no form.

The technician confirms with two taps: ✓ it enters the ERP, or ✏️ fix it before it enters. No training, no habit change. This is what sets iLEAN apart from an autonomous agent: no data crosses into the central system without a human reviewing it first.

This is the piece that makes all the others approvable. Every capture — the photographed order, the dictated incident, the supplier email — arrives as a short visual summary, and nothing crosses into the central system until a person confirms it. That is what separates iLEAN from an autonomous agent. Quality approves the flow because it keeps the last word.

See the full IRIS architecture →

Before and after

Data entry at the press today versus data confirmed with two taps

Aspect	Today	With iLEAN
Who sees a closure capture before the ERP	Nobody, or it is never captured	The technician at the press
Effort per record	A form at a terminal	A summary and two taps
Wrong color code or resin batch	Found by the bottler	Fixed before it enters
Order photos, voice and supplier emails	Three separate worlds	One confirmation screen
Corrections made at the press	Lost	Kept to improve extraction
Training for press technicians	A new ERP screen	None: confirm or fix

Unvalidated data contaminating the ERP → validated data in seconds, clean master data.

Impact estimate

An enabling piece for the closure plant — no standalone payback.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• No standalone payback: it is the enabling piece of the whole iLEAN matrix and its value is strategic, not directly monetizable. We say so instead of inventing a figure.
• Without it, the other pieces do not get signed off by quality or management, so the return of the matrix runs through this tablet even though none of it is booked here.
• From unvalidated data contaminating the ERP — or no data at all — to data captured and validated in seconds, by the person standing next to the press. No back-office cleanup, no weekly reconciliation between what happened and what was typed.
• And the master stays clean, which is what every certificate of conformity and every batch record sent to a bottler depends on. A dirty master is the fastest way to lose a bottler's trust in your certificates.

Strategic value · the piece that enables the whole matrix. The enabling piece for the whole iLEAN matrix — without it, the other pieces don't get signed off. Strategic value, not directly monetizable.

And the fair question from the production manager

“If a technician confirms everything, what is the AI saving?” — the assembly, not the decision. Pulling mold, cavity, color and resin batch out of a photo, a dictation or an email is an anchored task, where the best models drop below 1.5% error [1]; the technician only checks a summary that arrives already built. Two taps replace a form, and every correction is kept so the system learns where it fails. Over time he corrects less and confirms more. That is the sign the system is learning your presses.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.